// news · open-source2026-08-03source: llm-stats / computingforgeeks

Open weights now trail the closed frontier by only a few months, DeepSeek estimates

The gap is measured in months, not generations. DeepSeek estimates its open weights now trail the closed frontier by only a few months on many benchmarks, and Qwen 3.5 27B offers a strong dense alternative that's simpler to serve. For most workloads, the question is no longer whether open is good enough but whether the short lag matters.

A few months' lag is a different market than a few generations. When open weights trail the closed frontier by a short, closing interval on many benchmarks, the practical calculus flips: for the large share of workloads that don't need the absolute latest capability, an open model that is months behind at a fraction of the cost is the rational default, not the compromise.

Qwen 3.5 27B as a 'simpler to serve' dense alternative points at the other axis of the open market: operational fit. Not every deployment wants a 675-billion-parameter MoE; a capable dense model that runs on modest hardware is the right tool for many, and the open ecosystem now offers both the frontier-scale flagship and the easy-to-serve workhorse.

The pressure this sustains on the closed labs is structural. If the open gap keeps shrinking toward months, the closed premium must be justified by an ever-narrower window of leading-edge capability — which is precisely the dynamic behind the frontier price cuts. The open field's steady march is the background force setting the ceiling on what closed models can charge.

See our analysis →

LLM-Stats — AI updates today (August 2026) — latest model releases → · ComputingForGeeks — Open source LLM comparison table (2026) →